Configure Karpenter to mix spot and on-demand capacity using NodePool requirements and weighted NodePools

Steps

1. In a NodePool's spec.template.spec.requirements, add a requirement on key `karpenter.sh/capacity-type` with operator `In` and values `["spot", "on-demand"]` (optionally also `"reserved"`) so a single NodePool is allowed to launch either type.
2. To actually control the mix rather than just what's allowed, split into separate NodePools — one scoped to `karpenter.sh/capacity-type: spot`, another to `on-demand` — since a NodePool with both values in its requirements does not launch a proportional blend.
3. Use spec.weight on each NodePool (higher integer = tried first) to set priority ordering when a pod is compatible with more than one NodePool, e.g. `weight: 10` on the spot pool so it's preferred, with an unweighted (0) on-demand pool as fallback.
4. To enforce an actual ratio rather than a strict preference order, use `karpenter.sh/capacity-type` as a topology key in a pod's topologySpreadConstraints — this is the documented mechanism for guaranteeing a spot/on-demand split, not the NodePool requirements alone.
5. Add `minValues` to instance-category/instance-family/instance-type requirements on the spot-facing NodePool to force enough instance-type diversity for the Spot allocation strategy to reduce interruption risk.
6. Enable interruption handling at install time (`--set settings.interruptionQueue=<CLUSTER_NAME>`) so Karpenter proactively drains spot nodes ahead of the 2-minute EC2 reclamation notice, and confirm the resulting mix with `kubectl get nodes -L karpenter.sh/capacity-type`.